Who Benefits from Religious Attendance?

Heterogeneous Causal Effects on Well-being and Cooperation in New Zealand

Joseph A. Bulbulia

Victoria University of Wellington, New Zealand

Thanks

  • 76,409 individuals who have participated in the New Zealand Attitudes and Values Study since 2009.
  • Templeton Religion Trust
  • University of Auckland
  • Victoria University
  • Georgia State University
  • Don E. Davis
  • Ken Rice
  • Geoff Troughton
  • Chris G. Sibley & other collaborators
  • Grad Students in the EPIC Lap

Background

Overview

  1. New Zealand Attitudes and Values Study (NZAVS):

    • A nationally representative panel study, initiated in 2009.

    • Sample frame is drawn randomly from the NZ Electoral Roll.

    • Postal questionnaire (coverage; retention ~ 70-80%).

    • Large multidisciplinary research team (>60).

    • Focus on personality, social attitudes, values, religion, employment, prejudice \dots

    • Tracks nearly 1% of New Zealand’s adult population annually, with 76,409 participants to date.

  2. Religion’s Causal Effects on Cooperation

  3. Religion’s Causal Effects on Multi-Dimensional Flourishing

  4. Religion’s Causal Effects on Personality

What will we learn today?

Scope and Assumptions

  • Dis-engangling Causation from Correlation in Effects of Religious Service: Good-Life/ Cooperation
  • Focus on “Individual Differences”
  • Evolutionary Interests About How These Domains Are Related

Daily Counts of Processed NZAVS Surveys: Repeated Measures –> Causal Insights

Two Questions/ Two Domains

Eligibility

  • Responded to the time 10 (2018/19) New Zealand Attitudes Study
  • Information about religious service at baseline (yes or no).
  • Inverse Probability of Censoring Weights for Attrition
  • ‘Censored’ if responded during COVID lockdowns (NZAVS 10)
  • A total of 46,377 individuals met these criteria.

Method

  • Clearly stated causal estimands
  • Semi-parametric Machine Learning
  • Censoring models to recover what would have happened to population.
  • New Zealand Census Weights to for population inference
  • Cross-validation: Models always evaluated on unseen data

Context

Sample Demography

Table 1

Demographic statistics by religious denomination (NZAVS wave 10, approx~1.4% of NZ adult population).

Three Estimand Types

STUDY 1 Religious Service and the GOOD LIFE

1.1 Three Waves Good Life: ATE Causal Forests and Binary Exposure

Figure 4: ATE: Three-Wave: Binary Intervention

1.2 Good Life: Compare 3 waves Binary vs. 3 waves WEEKLY/ZER0

Figure 5: ATE: Three-wave Shift Intervention

1.3 Sensitivity Analysis: Compare 3 waves Religious Service WEEKLY/NULL with 3 waves Socialising Hours WEEKLY/NULL

Figure 6: ATE: Three-wave Soft Intervention

1.4 Good Life: Compare 3 waves WEEKLY/ZER0 vs 6 waves WEEKLY/ZER0

Figure 7: ATE: Six-wave Soft Intervention GAIN: +1

1.5 Good Life: Compare 6 waves WEEKLY/ZER0 vs 6 waves WEEKLY/NULL

Figure 8: ATE: Six-wave Soft Intervention LOSS: -1

1.6 Good Life: Compare 6 waves WEEKLY/NULL vs 6 waves ZERO/NULL

Figure 9: CATE RATE

SUMMARY STUDY 1: GOOD LIFE

  • Depends on which effect (single intervention vs five interventions/e.g. shift up/down)
  • Signals are detectable in binary intervention but weaker
  • Unlikely to be due to hours of socialising (an overly conservative sensitivity analysis)
  • Meaning and Purpose: Strongest Result
  • Sexual Satisfaction… (there’s a literature)
Figure 10: CATE rate

STUDY 2 Religious Service and COOPERATION

2.1 Three Waves Cooperation: ATE Causal Forests and Binary Exposure

Figure 11: ATE: Three-Wave: Binary Intervention

2.2 Three Waves Cooperation: Compare 3 waves Binary vs. 3 waves WEEKLY/ZER0

Figure 12: ATE: Three-wave Shift Intervention

2.3 Three Waves Cooperation: Compare 3 waves WEEKLY/ZER0 vs. 3 waves WEEKLY/NULL

Figure 13: ATE: Three-wave Soft Intervention

2.4 Six Waves Cooperation: Compare 3 waves WEEKLY/ZER0 vs 6 waves WEEKLY/NULL

Figure 14: ATE: Six-wave Soft Intervention WEEKLY/ZER0

2.5 Six Waves Cooperation: Compare 6 waves WEEKLY/NULL vs 6 waves ZERO/NULL

Figure 15: ATE: Six-wave Soft Intervention LOSS: -1

SUMMARY STUDY 2: COOPERATION

  • Again Inference depends on which effect we consider (single intervention vs five interventions/e.g. shift up/down)
  • Loss of religious service different to gain.
  • Not much signal in Social Support or Social Belonging
Figure 16: CATE RATE

STUDY 3 INDIVIDUAL DIFFERENCES (Heterogeneous Treatment Effects)

Policy Trees for The GOOD LIFE

Policy Tree 3.1: Forgiveness Outcome

Figure 17: Policy Trees for Forgiveness Outcome

Policy Tree 2: Gratitude Outcome

Figure 18: Policy Trees for Gratitude Outcome

Policy Tree 3: Hlth Fatigue Outcome

Figure 19: Policy Trees for Hlth Fatigue Outcome

Policy Tree 4: Meaning Sense Outcome

Figure 20: Policy Trees for Meaning Sense Outcome

Policy Tree 5: Neighbourhood Community Outcome

Figure 21: Policy Trees for Neighbourhood Community Outcome

Policy Tree 6: Self Esteem Outcome

Figure 22: Policy Trees for Self Esteem Outcome

Policy Tree for COOPERATION

Policy Tree 1: Donations

Figure 23: Policy Trees for log Charity Donate Outcome

SUMMARY

  • Ask a clear causal question
  • Loss and gains differ
  • Purpose in Life > Health Benefits
  • Signal in gain of Religious Service
  • Loss of religious service less impactful in short term
  • Longer term effects on volutneering
  • Social Belonging and Social Support: reliable signals are absent
  • Good Life Moderators:
    • usual suspects: Age/Personality/Health
    • but also: Fatigue/Housework/Income/Status/ Perfectionism
  • Cooperation
    • signal for Donations only
    • age/income/hours of housework.
  • Note: variables that rarely come up - Gender and Ethnicity
  • Noisy Signals (to exlude a null is not to endorse it)
  • Generalisations unclear
  • Big data representative for New Zealand
  • Repeated measures
  • Non parameteric machine-learning with cross-validation/ sample splitting
  • Individual Differences Reveal Importance of Baseline Demography and Health

Thank You!

References

Supplement 1

S.1 Heterogeneity Treatment Effect Decision Flow

This following flowchart shows the decision logic:

    START: For each model
             |
             v
    STEP 1: EXCLUSION CHECK
    Is RATE QINI < 0 (stat sig) OR RATE AUTOC < 0 (stat sig)?
             |
        +----+----+
        |         |
       YES       NO
        |         |
        v         v
    EXCLUDED    STEP 2: RATE SELECTION CHECK
    (Stop)      Is RATE QINI > 0 (stat sig) OR RATE AUTOC > 0 (stat sig)?
                         |
                    +----+----+
                    |         |
                   YES       NO
                    |         |
                    v         v
                SELECTED   UNCLEAR
                    |         |
                    v         v
                STEP 3: QINI CURVE ANALYSIS
                (Applied to all non-excluded models)
                    |         |
                    v         v
                Already     Check: Is QINI curve positive
                selected     at any spend level?
                            |
                       +----+----+
                       |         |
                      YES       NO
                       |         |
                       v         v
                   SELECTED   UNCLEAR
                             (final)